JPH03142691A - Table format document recognizing system - Google Patents

Abstract

PURPOSE:To perform table recognition in a short time and accurately by dividing image information into belt areas with prescribed width, and extracting a rule mark candidate from projection data in a direction intersecting orthogonally to a longitudinal direction. CONSTITUTION:A vertical rule mark candidate can be accurately extracted by providing an image scanner 1 and a graphic separation part 2, dividing read binary image information into, for example, the plural belt areas in a horizontal direction, extracting the projection data for each belt area in the direction intersecting orthogonally to the longitudinal direction, and extracting, for example, the rule mark candidate with narrow width in a vertical direction from the projection data, and a vertical rule mark can be recognized by connecting extracted rule mark candidates in neighboring belt areas. A horizontal rule mark can be extracted similarly. Also, when a table is the one in which a character string is partitioned with a null area, rule mark recognition can be easily performed by extracting a temporary rule mark candidate by retrieving the null area even when it is the table not partitioned with a line. In such a way, the character recognition rate and recognition speed of a character recognition device can be improved.

Description

[Detailed Description of the Invention] [Industrial Application Field] This invention reads a document containing a mixture of characters, tables, figures, etc. as document image information, and separates and recognizes individual information from this document image information. Concerning tabular document recognition methods.

[Conventional technology]

Conventionally, character recognition devices such as optical character recognition devices (OCR) read a document as image information, separate this image information into rows or columns and character groups, and further subdivide the parts that are considered to be characters for character recognition. However, when recognizing characters in a table, since the characters in the frame include ruled lines, the line spacing and character spacing are different, resulting in many misrecognitions. Since the structure of the table is not recognized, the problem is that the table itself loses its meaning.

In order to solve this problem, conventionally, as described in the Proceedings of the 1986 Institute of Electronics, Information and Communication Engineers Spring National Conference, pages 2-224, document images are scanned in the horizontal (or vertical) direction, and black pixels are Form a histogram of the numbers, and then, for the mountain part of the histogram, the section whose height is greater than or equal to the darkness value is set as the mountain width, and the one with the smaller width is extracted, and the corresponding area of the document image is divided into line segments. The line segment area is then scanned in the direction of the line segment, and the line segment is extracted with the end point being the part where white pixels are continuous over the darkness value.Then, the circumscribed rectangle of the table is calculated from the extracted line segment and column information is generated. It has been proposed to recognize the underlying structure.

[Problem to be solved by the invention]

However, in the conventional tabular document recognition method described above, when a document image is scanned horizontally or vertically, if the ruled lines are continuous in the horizontal or vertical direction, the ruled lines can be cut out. If there is a short vertical ruled line connecting two horizontal ruled lines, it cannot be detected, so in order to accurately recognize a finely divided table, it is necessary to recognize short ruled lines in the document image. It is necessary to divide each area into areas of a certain size and create a histogram by scanning each area in the horizontal and vertical directions, which increases the processing time for image recognition and creates ruled lines around the edges of the divided areas. In addition, it takes a long time to process when connecting the ruled line candidates extracted from each divided area, and furthermore, character groups are not separated by solid lines, dotted lines, etc., and virtual ruled lines are created by blank areas. There was an unresolved problem that table recognition could not be performed for tables divided by .

Therefore, this invention was made by focusing on the unresolved problems of the conventional example described above, and it is possible to perform table recognition accurately in a short time, and also to easily display even virtual ruled lines made of blank areas. The purpose is to provide a tabular document recognition method that can perform recognition.

[Means to solve the problem]

In order to achieve the above object, the tabular document recognition method according to claim (1) reads a tabular document containing a mixture of characters and ruled lines with an image reading device, and uses the image information in the image reading device to read the tabular document in which characters and ruled lines are mixed. This is a tabular document recognition method in which ruled lines are cut out and character information in areas that are bottomed by the ruled lines is sent to a character recognition device. projection data forming means for forming projection data in which is projected in a direction perpendicular to the longitudinal direction;
ruled line candidate extraction means for extracting ruled line candidates from the projection data formed by the projection data forming means; ruled line recognition means for recognizing ruled lines by connecting the ruled line candidates extracted by the ruled line candidate extraction means between adjacent strip areas; It is characterized by having the following.

Further, in the tabular document recognition method according to claim (2), in the claim (1), when there is a blank area between character strings, the ruled line candidate extracting means extracts a virtual ruled line candidate from the blank area. It is characterized by being configured to do so.

[Effect]

In the tabular document recognition method according to claim (1), the binarized image information read by the image reading device is divided into, for example, a plurality of strip-shaped regions in the horizontal direction, and each of these strip-shaped regions is divided orthogonally to its longitudinal direction. Take the projection data in the direction of
For example, by extracting narrow vertical ruled line candidates from this projection data, vertical ruled line candidates can be accurately extracted, and by connecting these extracted ruled line candidates between adjacent strip areas, vertical ruled line candidates can be can be recognized.

In addition, in order to extract horizontal ruled lines, the binarized image information is divided into multiple strip-shaped regions in the vertical direction, horizontal ruled line candidates are extracted from projection data perpendicular to the longitudinal direction, and ruled line candidates for each strip-shaped region are extracted. By concatenating , horizontal ruled lines can be extracted. When extracting this horizontal ruled line, by deleting the vertical ruled line extracted by the process from the binarized image information in advance, the time for recognizing the ruled line can be shortened.

In addition, in the tabular document recognition method according to claim (2), when the character string is a table divided by blank areas, by searching the blank areas and extracting virtual ruled line candidates, line Even if the table is not divided into sections, it is possible to easily recognize ruled lines.

〔Example〕

Embodiments of the present invention will be described below based on the drawings.

FIG. 1 is a block diagram showing a schematic configuration of the present invention.

In the figure, l is an image scanner, which is controlled by a processing device 4 composed of a personal computer or the like, and as shown in FIG. 2, vertical ruled lines Rv and horizontal ruled lines RH
A tabular document in which arbitrary characters are drawn in a table T formed by is read, and this is binarized into black and white pixels at a predetermined level to output binarized image information.

The binarized image information from the image scanner 1 is sent to the figure separator 2, where it is divided into table information and character information separated by the table information. ) 3, the optical character recognition device 3 performs character recognition, and the recognized character data is input to the processing device 4.

The figure separator 2 has at least an interface circuit 2a, an arithmetic processing unit 2b, and a storage device 2C including a dot-compatible image memory, and when the binary image information from the image scanner 1 is input, the figure separator 2 converts the binary image information into The converted image information is stored in the image memory of the storage device 2C, and then the processing shown in FIG. 3 is executed by the arithmetic processing device 2b. That is, first, step 2 executes vertical ruled line extraction processing to extract vertical ruled lines Rv, then moves to step 2, executes horizontal ruled line extraction processing to extract horizontal ruled lines Rg, and then moves to step 2, which extracts horizontal ruled lines Rv. A virtual ruled line extraction process is executed to extract a virtual ruled line R, which is a blank area corresponding to the ruled line, and then the process proceeds to step 2, where vertical ruled lines RV, extracted in each process are extracted.
The image information in the field surrounded by the horizontal ruled line R and the virtual ruled line R7 is sent to the optical character reading device 3.

Here, in the vertical ruled line extraction process, as shown in FIG. 4, the binarized image information stored in the image memory in step A plurality of N band-shaped regions B X I-B X extending in the X direction)
A predetermined width W in the vertical direction is read out sequentially in the horizontal direction so as to divide it into As shown in FIG. 7(b), vertical projection data of the logical value "l" which is a black pixel is formed, and then the process proceeds to step @, where a vertical ruled line candidate is extracted from this projection data.This vertical ruled line candidate To extract the data, the projection data is binarized again using a preset threshold, and this binarized data is
Search for consecutive parts of ``l'', and since ruled lines are usually narrow, if the continuous width of ``°1'' is less than a preset width, this is extracted as a vertical ruled line candidate, and the others are used as character candidates. Extract as.

Next, the process moves to step (2), and a ruled line candidate re-extraction process is executed to extract vertical ruled line candidates included in the character candidates. That is, as shown in FIG. 7(a), when a certain strip area BX includes a horizontal ruled line R□, its projection data includes a horizontal ruled line RH as shown in FIG. 7(b). Since the logical value "1" is increased by the amount represented by the logical value "1", even if binarization is performed again at the predetermined threshold value T shown by the dotted line in FIG. Therefore, in the ruled line candidate re-extraction process, the maximum value in the width interval where the logical value "1" is continuous (if it includes a ruled line, the ruled line By performing binarization again using the average value within the width section (which usually has a high peak) and the average value within the width section, vertical ruled line candidates included in the character candidates can be determined, as shown in Figure 7 (C). can be extracted.

Next, proceeding to step [phase], each band-shaped region BX1
~BX, it is determined whether or not extraction of ruled line candidates has been completed, and if there is a band-shaped area for which ruled line candidate extraction has not been completed, the process moves to step [phase] and the successive addresses of the next image memory are Since the setting is made so that the strip area BX11 is set, return to the step
When the extraction of ruled line candidates for t~BX is completed, the process moves to step [phase] and the extraction of ruled line candidates for each strip area BX, ~BX is completed.
The vertical ruled line candidates extracted in , are vertically connected to form a vertical ruled line candidate sequence. To form this vertical ruled line candidate row, from each ruled line candidate group in the vertical direction (
This is done by extracting vertical ruled line candidates that overlap in the Y direction) and connecting them.

Next, in step @, the vertical ruled line candidate row is projected in a direction perpendicular to its extension direction, that is, in the X direction to obtain projection data, and when this projection data is continuous at a certain level or more, it is recognized as a ruled line. , the vertical ruled lines are determined by determining the start and end coordinates from the projection data, and then the process is terminated and the process returns to the process of FIG. 3.

In the horizontal ruled line extraction process, as shown in FIG. Binarized image information is created and stored in the image memory.

Next, the process moves to step @, and for the erased binarized image information, the strip area B in the vertical ruled line candidate extraction process is
Instead of X, ~B

The erased binarized image information is read out vertically by a predetermined width and projected in the X direction perpendicular to the longitudinal direction (Y direction) to create projection data.

Next, in step 0, horizontal ruled line candidates are extracted by performing the same process as the vertical ruled line candidate extraction process based on the projection data.

Next, the process moves to step [phase] to determine whether extraction of horizontal ruled line candidates has been completed for all the vertical strip areas BY, -BY, and if the extraction has not been completed, the process moves to step @. After setting the successive address corresponding to the next vertical strip area B Y j + r, return to step @, and when the extraction of horizontal ruled line candidates for all vertical strip areas BY1 to BY is completed, proceed to step [phase]. Transition.

In this step [phase], a horizontal ruled line candidate row is created by extracting and concatenating horizontal ruled line candidates in adjacent vertical strip areas, and then the process moves to step O to project the horizontal ruled line candidate row in the X direction. By doing this, the horizontal ruled lines are determined.

Furthermore, as shown in FIG. 6, in the virtual ruled line extraction process, first, in step [phase], as shown in FIG. Then, the inside of the rectangular outermost ruled line frame F circumscribing the outermost ruled line of the horizontal ruled line RM is set as a virtual ruled line processing area, and then the process moves to step @, and the vertical ruled line stored in the image memory is erased and binarized. Horizontal rule L9 from image information
．． R, is deleted to form binarized image information as shown in FIG. into strip regions BY, -BY, and these strip regions B
Projection data in the X direction is formed for each of Y, ~BYA, and then the process moves to step [phase], where an area representing a logical value "0" is extracted as a virtual ruled line candidate from the projection data.As a result, for example, As shown in FIG. 1O(a), virtual ruled line candidates for blank areas can be extracted from each band-shaped area BYi-, , BYe, and BYt-t.

Next, the process moves to step [phase] to determine whether extraction of virtual ruled line candidates has been completed for all vertical strip areas BY, -BY, and there is a vertical strip area for which extraction of virtual ruled line candidates has not been completed. Sometimes, the process moves to step [phase] and successive addresses for the image memory are transferred to the next strip area BY,...
Update to the address corresponding to , then return to step 0,
When the extraction of virtual ruled line candidates is completed for all vertical strip areas B Y + to BYw, the process moves to step (2).

In this step (2), the virtual ruled line candidates extracted in the step [phase] include all the blank areas, so the difference in the size of the candidates becomes significant. For example, in the strip area BY□ in FIG. 10(a) Then, since all of them become virtual ruled line candidates, the virtual ruled line candidates of this strip area BYt and the adjacent strip areas BYe and BY. The numbers of virtual ruled line candidates CA r and CAZ are calculated, the numbers of both virtual ruled line candidates are compared, and the virtual ruled line candidates are divided according to the larger virtual ruled line candidate to create virtual ruled line division candidates CA.

and select CAz'.

Next, the process moves to step [phase], and among all virtual ruled line candidates including the extracted virtual ruled line division candidates, virtual ruled line candidates that are horizontally connected to each other in adjacent strip wI areas are searched and connected, and the virtual ruled line candidates are searched and connected. Form candidate columns.

Next, the process moves to step 0, in which virtual ruled line candidate columns that overlap with the vertical ruled line Rv and horizontal ruled line R9 determined in the above-described ruled line candidate extraction process are deleted from the virtual ruled line candidate columns, and virtual ruled line candidate columns are deleted for the remaining virtual ruled line candidate columns. After determining this, the process returns to the process shown in FIG. In determining this virtual ruled line, since the virtual ruled line candidate row is wider than the ruled line, the midpoint of the virtual ruled line candidate at the start and end of the virtual ruled line candidate row is used as a representative point,
By determining the coordinates of this representative point, a line segment connecting both representative points is determined as a virtual ruled line.

In addition, when determining the virtual ruled lines, please refer to Fig. 11(a).
As shown in the figure, there are cases where multiple narrow virtual ruled line candidates CA + z and CA13 are connected to the wide virtual ruled line candidate CA, and in this case, the wide virtual ruled line candidate CA
+ Midpoint of + and narrow virtual ruled line candidate CA2. If the midpoint of is selected as the representative point, virtual ruled lines will be extracted incorrectly, so when a virtual ruled line candidate string is formed by connecting wide virtual ruled line candidates and narrow virtual ruled line candidates, , 11th
As shown in Figure (B), the midpoint of one of the narrow virtual ruled line candidates is selected as the only representative point to determine the virtual ruled line, and the remaining narrow virtual ruled line candidates are The boundary position touching the wide ruled line candidate is determined as the end point or starting point coordinates of the virtual ruled line.

In this way, when the recognition of the tabular format is completed by determining all the ruled lines and virtual ruled lines, the process moves to step (2) in FIG. Output to character recognition device 3.

Therefore, the optical character recognition device 3 processes 1 m of characters for each column of the tabular document, which improves the character recognition rate and recognizes the back structure. It is now possible to specify only letters or only numbers for each item in a row or column of a document, and the number of selected characters can be limited, improving recognition rate and speed. . Furthermore, by converting the characters recognized by the optical character recognition device 3 into control codes and inputting them into the arithmetic processing device 4 composed of a personal computer or the like, names and addresses in address books, telephone directories, etc.
When creating a database of telephone numbers, etc., it is only necessary to read these documents with an image scanner, and the database can be created easily.

Note that in the above embodiment, the case where the image scanner 1 is applied as the image reading device has been described.
The present invention is not limited to this, and any image reading device such as a television camera, solid-state imaging device, etc. can be applied.

Furthermore, in the above embodiment, the case where the ruled line of the tabular document is a solid line has been explained, but the invention is not limited to this, and even when the ruled line is a broken line, a chain line, etc., the peak of the projection data is a solid line. The ruled lines can be easily recognized because they are only smaller than in the case where the ruled lines are displayed.

Furthermore, in the above embodiment, the image scanner 1
The case where the reading direction of the image scanner 1 matches the ruled line direction of the tabular document has been described; When determining the start and end coordinates of vertical ruled lines Rv, horizontal ruled lines RH, and virtual ruled lines IIRI, projection data in a direction perpendicular to the projection direction used when extracting ruled line candidates from the ruled line candidate array is taken, and the starting and ending points are determined using projection data. By finding the coordinates and the equation of the ruled line candidate row,
Starting point and ending point coordinates can be determined. However, as for the virtual ruled line candidates, since there are no real ruled lines as described above, an equation of a straight line connecting the midpoints of the start and end points of the virtual ruled line candidate row, which are representative points, is used.

〔Effect of the invention〕

As explained above, in the tabular document recognition method according to claim (1), the vertical ruled lines and the horizontal ruled lines forming the table of the tabular document are binarized by dividing them into strip-shaped areas perpendicular to both. Since ruled line candidates are extracted based on projection data obtained by projecting image information in a direction perpendicular to its longitudinal direction, the number of divisions of binarized image information can be reduced, and the recognition speed of tabular documents can be increased. can be improved,
Moreover, since the back structure is recognized and character recognition is performed for each column of this back structure, it is possible to specify only letters or only numbers for each item in a row or column, and the character recognition device By being able to limit the number of selected characters, it is possible to improve the character recognition rate and recognition speed of the literary journal m device, and it is also possible to recognize tabular formats, so it is possible to create a database for each predetermined item from tabular documents. In this case, an advantage can be obtained that the tabular document can be easily created simply by reading it with a character reading device.

In addition to the effect of claim (1), the tabular document recognition method according to claim (2) also provides a table format that is not divided by ruled lines but is divided by virtual ruled lines formed by blank areas. Even if the character is in a different format, it can be recognized as having ruled lines, so that the character recognition rate and recognition speed of the character recognition device can be further improved.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing a schematic configuration of the present invention, FIG. 2 is an explanatory diagram showing a tabular document, FIGS.
Figures (a), (b), and (C) are explanatory diagrams showing a horizontal band-shaped area, its vertical projection data, and re-binarized data, respectively, to provide a detailed explanation of the present invention, and Fig. 8 shows a virtual ruled line processing area. FIG. 9 is an explanatory diagram showing erased binarized information, FIG. 10 (a) and operation) and FIG. 11 (a) and (
b) is an explanatory diagram for explaining the virtual ruled line extraction process. In the figure, 1 is an image scanner (character reading device), 2 is a graphic separation unit, 3 is an optical character recognition device, 4 is an arithmetic processing unit,
D is a tabular document, Rv is a vertical ruled line, and RN is a horizontal ruled line.

Claims (2)

[Claims] (1) Read a tabular document containing a mixture of characters and ruled lines with an image reading device, cut out the ruled lines in the image information from the image reading device, and send the character information in the area defined by the ruled lines to a character recognition device. A tabular document recognition method configured to send the image information into strip-shaped areas of a predetermined width;
projection data forming means for forming projection data obtained by projecting each strip area in a direction orthogonal to the longitudinal direction thereof; a ruled line candidate extracting means for extracting ruled line candidates from the projection data formed by the projection data forming means; and the ruled line candidates. 1. A tabular document recognition method, comprising a ruled line determining means for determining a ruled line by connecting ruled line candidate areas extracted by the extracting means between adjacent strip areas. (2) The tabular document recognition method according to claim 1, wherein the ruled line candidate extracting means is configured to extract virtual ruled line candidates from the blank area when there is a blank area between character strings.